报告内容简介: It is important to identify seismic events for the study of earthquake early warning. With the development of convolutional neural network (CNN), people have made substantial progress on image recognition. Using the theory of convolutional neural networks, a network structure can be obtained to identify tele-seismic events by learning the waveform of the tele-seismic events. We have analyzed different types of samples, including changes in sample shape, changes in information contained in the sample and changes in sample input format. Our goal is to find the best sample type and tele-seismic events that can be more accurately identified. The samples contain not only actual data but also theoretical data. Both types of data are trained and learned. In addition, impacts of different optimization algorithms have also been studied in our research. The most efficient optimization algorithm is selected. When the new sample type is entered into the post-learning network, the convolutional neural network can well recognize tele-seismic events and meet expectations.
